Author: Drexel,Michael
Date: 13:22:06 09/29/03
Go up one level in this thread
On September 29, 2003 at 15:35:23, Jeroen van Dorp wrote: >On September 29, 2003 at 14:36:52, Drexel,Michael wrote: > >>>So you basically want an on the fly analysis which has no value to you as well, >> >>Did I say it has no value for me? >>No > > >Of course not. It is an inevitable conclusion from your own words. In a test >suite you know the answer. But knowing the answer is not only displaying the >answer, but also knowing why. You only tested if the engines could display the >answer, and added that you didn't trust other answers anyhow, maybe just these, >because you knew the answers. > >If you don't trust it in actual analysis, why trust it in a test suite? Just >because you know the answers, and the engine happened to reproduce that move as >one of the first candidates? I trust it because there are many positions and not only a single position. And yes, I am happy if it reproduces the key moves as one of the first candidate moves more often than another engine does. > > > >>It helps me a lot to initiate my own thinking process. >>If I agree with the engine move I can go on. If I dont agree for some reason I >>have to find and check the alternatives. > > >So why the testing with a test suite? Did it test the ability of an engine to >deliver a move you agreed with? A test suite is certainly of no value with that. > > >>Certainly an engine can find the right move for the wrong reason and discard it >>later on. >>However if a program finds the right moves on average earlier than another >>program then it is also more likely that it shows convincing score and pv >>earlier because moves at the top of the list are calculated longer. > > >So finding the right move for the wrong reason is an indication of strenght over >other engines? You cannot prove that. What are you talking about? In most of the cases an engine finds the right move for the right reason. This is just a question of probability. There are cases when it finds the right move for the wrong reason, but that is true for all the other programs as well. > >Finding a move with a wrong calculation is no guarantee at all that analysis >eventually will turn out okay because there's more time spent on that PV. It >only supposes the PV is wrong. When using the maximal time it could well be that >it discards the "right move" never to return to the right move again. How often do I have to repeat it? It is not important for me whether the PV is wrong or not or whether it discards the move or not later on and never returns again. > >You wanted to find out which engine would cater you best in on-the-fly analysis. >You suggest that breaking off a search when it happens to display the correct >move in a test suite gives a good indication of that. > Yes Eventually I do exactly the same during interactive Analysis. >The only indication of the best "on te fly" analysis is an engine finding the >right move with the correct assessment in the shortes time avalable and not >changing its analysis and PV for the longest possible time. It is my task to find the right move with the correct assessment. Michael > >That's why Mike is right, and it's even for _your_ purposes best to run the >tests as proposed. Your results seem to have little value, both for comparing >"raw" engine strenght as for "on-the-fly" analysis strenght. > >J.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.